On Fri, 4 May 2007 08:24:34 -0300
"Arnaldo Carvalho de Melo" <[EMAIL PROTECTED]> wrote:
> On 5/4/07, Eric Dumazet <[EMAIL PROTECTED]> wrote:
> > Hum, maybe not enough tea this morning, but I always thought spinlock_t size
> > was 4 bytes
>
> Well, unfolding sk_buff_head to see the details:
>
On 5/4/07, Eric Dumazet <[EMAIL PROTECTED]> wrote:
Arnaldo Carvalho de Melo a écrit :
>
> [EMAIL PROTECTED] linux-2.6]$ pahole -C sk_buff_head net/core/sock.o
> struct sk_buff_head {
> struct sk_buff * next; /* 0 8 */
> struct sk_buff * prev
On 5/4/07, David Miller <[EMAIL PROTECTED]> wrote:
sk_buff_head is due for being killed from the whole tree. Nobody
really needs the qlen, few things really need the lock, and those that
do can define their own as needed :-)
I've got out of tree research code that uses the qlen quite
significa
Arnaldo Carvalho de Melo a écrit :
[EMAIL PROTECTED] linux-2.6]$ pahole -C sk_buff_head net/core/sock.o
struct sk_buff_head {
struct sk_buff * next; /* 0 8 */
struct sk_buff * prev; /* 8 8 */
__u32
From: Arnaldo Carvalho de Melo <[EMAIL PROTECTED]>
Date: Fri, 4 May 2007 01:11:01 -0300
>
> This saves 8 bytes out of struct sock in 64bit arches, tested on x86_64.
Thanks I'll apply this.
> Now struct sock has this layout, where we waste more 20 bytes, in four 4 byte
> paddings in struct
Hi David,
This saves 8 bytes out of struct sock in 64bit arches, tested on x86_64.
[EMAIL PROTECTED] linux-2.6]$ codiff sock.o.before net/core/sock.o
/home/acme/git/linux-2.6/net/core/sock.c:
struct sock | -8
1 struct changed
[EMAIL PROTECTED] linux-2.6]$
Now struct sock has this la
